Very interesting application, but I think they'll have a hard time attempting to approximate the amount of information a human competitor would have (assuming they want to limit the machine's knowledge to better represent human memory).
I would rather have them give it access to Google's entire cache, since the real feat is processing and understanding the questions and then giving an appropriate answer. In theory, if the machine has 100% question recognition and can research an answer as fast as Google serves its pages, it should wipe the floor with the human contestants.
I do hope they find a way to keep the contest competitive and interesting.
The problem that Google solves and the problem that Watson solves are completely different. Sure, there is some overlap, because you might be able to type a Jeopardy question (answer?) into Google, and, as a human, quickly and easily find the correct answer. This is a side effect of very well-tuned retrieval algorithms that search for just the right words in your query, take into account user click-throughs for similar queries, and rank information based on billions of documents and eleven years of experience.
The Jeopardy problem is concerned natural language processing and knowledge representation as opposed to a tumult of data. I would wager that Watson's internal database is structured more like a single ontology than a highly distributed grid. The challenge here is mapping key parts of the question to this ontology and returning a single, highly precise result.
I would rather have them give it access to Google's entire cache, since the real feat is processing and understanding the questions and then giving an appropriate answer. In theory, if the machine has 100% question recognition and can research an answer as fast as Google serves its pages, it should wipe the floor with the human contestants.
I do hope they find a way to keep the contest competitive and interesting.